How do we figure out which teams are strongest and most likely to win any given match? Our model considers two main approaches. Goal-based models simulate how many goals each side will score, then work out the result from there. Outcome-based models skip the goals and directly estimate the probability of a win, draw, or loss. Both have real strengths, and understanding how they work is key to understanding the probabilities we publish.
How we make predictions
How We Rate Team Strength
Summary + sample · full document is 1,977 words
Summary
Sample
The two modelling traditions
Goal-process models treat scoring as a stochastic process — typically Poisson — where each team has a latent attack/defence rate. Results are simulated by sampling goals.
Outcome-prediction models skip the goal process and directly output P(home / draw / away) — or P(home win by ≥ k) and P(total ≥ n) for derived targets — from features. Discriminative classifiers, ratings systems, and regression models all sit here.
The two traditions complement each other. Goal models give you a coherent joint distribution over scores (good for derived targets: BTTS, exact-score grid, goal-differential lines), at the cost of strong distributional assumptions. Discriminative models often beat goal models on raw 1X2 log-loss but can't produce a coherent joint distribution for derived targets.
Full document
Pro Pass
Want the full document?
How We Rate Team Strength runs 1,977 words. The Pass unlocks this document and every research note in full, plus per-fixture probabilities, the four-model comparison, and per-fixture tactical analysis.
Every forecast graded against the real result, scored on 987 matches since 2014. See the scorecard.
24h money-back, no questions asked·No subscription, no auto-renewal·Access through 31 Dec 2026. See refund policy.